Boost 最大流算法无法编译.错误:形成对 void 的引用 [英] Boost max flow algorithms do not compile. error: forming reference to void

问题描述

Boost 提供了三种不同的算法来寻找有向图中的最大流:boykov_kolmogorov、edmonds_karp 和 push_relabel.它们都有命名和非命名参数版本.他们使用的参数集也非常相似.尽管如此，使用相同的参数，这些算法中的一些可以编译，而另一些则不能.

push_relabel 可以很好地编译命名和非命名版本:

 使用 Graph =boost::adjacency_list<boost::vecS, boost::vecS, boost::directedS,VertexProperty、EdgeProperty>;auto props = boost::capacity_map(容量).residual_capacity_map(residual_capacity).reverse_edge_map(reverse_edge_map).vertex_index_map(vertex_index_map).color_map(color_map).predecessor_map(predcessor_map).distance_map(distance_map);boost::push_relabel_max_flow(g, s, t, props);boost::push_relabel_max_flow(g, s, t, 容量, 剩余容量,reverse_edge_map, vertex_index_map);

boykov_kolmogorov 使用未命名版本编译:

 boost::boykov_kolmogorov_max_flow(g, 容量, 剩余容量,reverse_edge_map,vertex_index_map, s, t);

但命名版本失败:

 boost::boykov_kolmogorov_max_flow(g, s, t, props);

<块引用>

/celibs/boost_1_73_0/boost/graph/detail/adjacency_list.hpp:2768:17:错误:形成对void的引用

edmonds_karp 失败，命名和非命名版本均出现相同错误:

boost::edmonds_karp_max_flow(g, s, t, props);boost::edmonds_karp_max_flow(g，s，t，容量，residual_capacity，reverse_edge_map，color_map, predcessor_map);

<块引用>

/celibs/boost_1_73_0/boost/concept_check.hpp:147:9: 错误:使用已删除的函数

完整示例在这里:https://godbolt.org/z/dvjfec

我是否以错误的方式传递参数?如何正确传递它们?

谢谢！

解决方案

这确实似乎是一个错误.

当没有定义内部 edge_capacity_t 属性(室内属性).

定义一个会使问题消失.但是，我们可以检查它是否始终优先于通过命名参数提供的参数:

struct Oops {};使用 EdgeProperty = boost::property;

导致编译问题，提示选择了错误的属性映射.

我找不到造成这种行为的明显原因 - 所有其他命名参数的行为都符合预期，并且以非常相似的方式声明(该过程由宏自动执行).我认为会涉及一些非常微妙的事情(例如名称冲突或 ADL 事故?).

这是对我有用的代码:

Live On Wandbox(注意显然不能成功运行，因为它不满足任何不变量)

#define BOOST_ALLOW_DEPRECATED_HEADERS#include #include #include #include <boost/graph/edmonds_karp_max_flow.hpp>#include #include #include int main() {struct VertexProperty final {};//struct EdgeProperty final {};使用 EdgeProperty = boost::property;使用图 =boost::adjacency_list<boost::vecS, boost::vecS, boost::directedS,VertexProperty、EdgeProperty>;使用 Edge = boost::graph_traits::edge_descriptor;使用 Vertex = boost::graph_traits::vertex_descriptor;自动 g = 图{}；汽车 s ​​= 顶点{};自动 t = 顶点{}；自动残留容量映射 = std::vector{};auto reverseEdgeMap = std::vector{};auto colorMap = std::vector{};auto predcessorMap = std::vector{};auto distanceMap = std::vector{};自动顶点索引图 =boost::make_function_property_map([](Vertex) { return 0; });自动 edge_index_map =boost::make_function_property_map([](Edge) { return 0; });//自动容量 = boost::make_function_property_map( [](Edge) { return 0; });自动容量 = boost::get(boost::edge_capacity, g);自动剩余容量 = boost::make_iterator_property_map(残差图.begin(), edge_index_map);auto reverse_edge_map = boost::make_iterator_property_map(reverseEdgeMap.begin(), edge_index_map);自动 color_map =boost::make_iterator_property_map(colorMap.begin(), vertex_index_map);自动 predcessor_map = boost::make_iterator_property_map(predcessorMap.begin(), vertex_index_map);auto distance_map = boost::make_iterator_property_map(distanceMap.begin(),顶点索引图)；auto props = boost::capacity_map(容量).residual_capacity_map(residual_capacity).reverse_edge_map(reverse_edge_map).vertex_index_map(vertex_index_map).color_map(color_map).predecessor_map(predcessor_map).distance_map(distance_map);boost::push_relabel_max_flow(g, s, t, props);boost::push_relabel_max_flow(g, s, t, 容量, 剩余容量,reverse_edge_map, vertex_index_map);boost::boykov_kolmogorov_max_flow(g，容量，residual_capacity，reverse_edge_map, vertex_index_map, s, t);boost::boykov_kolmogorov_max_flow(g, s, t, props);boost::edmonds_karp_max_flow(g, s, t, props);boost::edmonds_karp_max_flow(g, s, t, 容量, 剩余容量,reverse_edge_map, color_map, predcessor_map);}

如您所见，所有算法调用现在都可以编译.

Boost provides three different algorithms for finding max flow in directed graphs: boykov_kolmogorov, edmonds_karp and push_relabel. All of them have named and non-named parameter versions. Parameter sets they use are also very similar. Despite that, with same parameters some of these algorithms compile and some of them do not.

push_relabel compiles nicely with both named and non-named version:

  using Graph =
    boost::adjacency_list<boost::vecS, boost::vecS, boost::directedS,
                        VertexProperty, EdgeProperty>;
  auto props = boost::capacity_map(capacity)
               .residual_capacity_map(residual_capacity)
               .reverse_edge_map(reverse_edge_map)
               .vertex_index_map(vertex_index_map)
               .color_map(color_map)
               .predecessor_map(predcessor_map)
               .distance_map(distance_map);
  boost::push_relabel_max_flow(g, s, t, props);
  boost::push_relabel_max_flow(g, s, t, capacity, residual_capacity,
                   reverse_edge_map, vertex_index_map);

boykov_kolmogorov compiles with non-named version:

  boost::boykov_kolmogorov_max_flow(g, capacity, residual_capacity,
                                    reverse_edge_map,
                                    vertex_index_map, s, t);

But fails with named version:

  boost::boykov_kolmogorov_max_flow(g, s, t, props);

/celibs/boost_1_73_0/boost/graph/detail/adjacency_list.hpp:2768:17: error: forming reference to void

edmonds_karp fails with both named and non-named versions with same error:

boost::edmonds_karp_max_flow(g, s, t, props);
boost::edmonds_karp_max_flow(g, s, t, capacity, residual_capacity, reverse_edge_map,
                          color_map, predcessor_map);

/celibs/boost_1_73_0/boost/concept_check.hpp:147:9: error: use of deleted function

Full example is here: https://godbolt.org/z/dvjfec

Do I pass parameters in incorrect way? How to pass them correctly?

Thanks!

解决方案

This indeed seems to be a bug.

It appears that the choose_const_pmap for edge_capacity fails when there is no interior edge_capacity_t property defined (Interior Properties).

Defining one makes the problem go away. However, we can check that it always takes precedence over the one provided via named paramaters:

struct Oops {};
using EdgeProperty = boost::property<boost::edge_capacity_t, Oops>;

Leads to compilation problems, suggesting that the wrong property map is selected.

I could not find an obvious cause for this behaviour - all the other named arguments behave as expected, and are declared in very similar fashion (the process is automated by macro). I assume there will be something very subtle involved (like a name collision or ADL mishap?).

Here's the code that works for me:

Live On Wandbox (note obviously can't run successfully, because it doesn't satisfy any invariants)

#define BOOST_ALLOW_DEPRECATED_HEADERS
#include <boost/config.hpp>
#include <boost/graph/adjacency_list.hpp>
#include <boost/graph/boykov_kolmogorov_max_flow.hpp>
#include <boost/graph/edmonds_karp_max_flow.hpp>
#include <boost/graph/graph_utility.hpp>
#include <boost/graph/push_relabel_max_flow.hpp>
#include <boost/property_map/function_property_map.hpp>

int main() {
    struct VertexProperty final {};
    // struct EdgeProperty final {};
    using EdgeProperty = boost::property<boost::edge_capacity_t, int>;
    using Graph =
        boost::adjacency_list<boost::vecS, boost::vecS, boost::directedS,
                              VertexProperty, EdgeProperty>;
    using Edge = boost::graph_traits<Graph>::edge_descriptor;
    using Vertex = boost::graph_traits<Graph>::vertex_descriptor;
    auto g = Graph{};
    auto s = Vertex{};
    auto t = Vertex{};

    auto residualCapacityMap = std::vector<int>{};
    auto reverseEdgeMap = std::vector<Edge>{};
    auto colorMap = std::vector<boost::default_color_type>{};
    auto predcessorMap = std::vector<Edge>{};
    auto distanceMap = std::vector<int>{};

    auto vertex_index_map =
        boost::make_function_property_map<Vertex>([](Vertex) { return 0; });
    auto edge_index_map =
        boost::make_function_property_map<Edge>([](Edge) { return 0; });
    // auto capacity = boost::make_function_property_map<Edge>( [](Edge) { return 0; });
    auto capacity = boost::get(boost::edge_capacity, g);
    auto residual_capacity = boost::make_iterator_property_map(
        residualCapacityMap.begin(), edge_index_map);
    auto reverse_edge_map = boost::make_iterator_property_map(
        reverseEdgeMap.begin(), edge_index_map);
    auto color_map =
        boost::make_iterator_property_map(colorMap.begin(), vertex_index_map);
    auto predcessor_map = boost::make_iterator_property_map(
        predcessorMap.begin(), vertex_index_map);
    auto distance_map = boost::make_iterator_property_map(distanceMap.begin(),
                                                          vertex_index_map);

    auto props = boost::capacity_map(capacity)
                     .residual_capacity_map(residual_capacity)
                     .reverse_edge_map(reverse_edge_map)
                     .vertex_index_map(vertex_index_map)
                     .color_map(color_map)
                     .predecessor_map(predcessor_map)
                     .distance_map(distance_map);

    boost::push_relabel_max_flow(g, s, t, props);
    boost::push_relabel_max_flow(g, s, t, capacity, residual_capacity,
                                 reverse_edge_map, vertex_index_map);

    boost::boykov_kolmogorov_max_flow(g, capacity, residual_capacity,
                                      reverse_edge_map, vertex_index_map, s, t);
    boost::boykov_kolmogorov_max_flow(g, s, t, props);

    boost::edmonds_karp_max_flow(g, s, t, props);
    boost::edmonds_karp_max_flow(g, s, t, capacity, residual_capacity,
                                 reverse_edge_map, color_map, predcessor_map);
}

As you can see all of the algorithm invocations now compile.

这篇关于Boost 最大流算法无法编译.错误:形成对 void 的引用的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！